Internal Migration in Kenya: Implications for Maternal and Child Healthcare UtilizationbyJulia M. PorthA dissertation submitted in partial fulfillmentof the requirements for the degree ofDoctor of Philosophy(Epidemiological Science)in the University of Michigan2021Doctoral Committee:Professor Matthew L. Boulton, ChairProfessor Thomas M. BraunAssociate Professor Nancy L. FleischerAssociate Professor Cheryl A. MoyerDr. Martin K. Mutua, African Population and Health Research CenterResearch Assistant Professor Abram L. Wagner

Julia M. Porthjm[email protected] iD: 0000-0001-6407-5579 Julia M. Porth 2021

DedicationTo my family, who always encourage me to chase my dreams.ii

AcknowledgementsThis dissertation would not be possible without the incredible support I have receivedover the past few years. First, I am grateful to my dissertation chair, Dr. Matthew Boulton, forhis mentorship and guidance throughout my MPH and doctoral studies. Matt, your advice,perspective, and encouragement are deeply appreciated and have made me a better researcher. Iam additionally thankful for my wonderful dissertation committee, Dr. Thomas Braun, Dr.Nancy Fleischer, Dr. Cheryl Moyer, Dr. Martin Mutua, and Dr. Abram Wagner. Working witheach of you has been a pleasure and I feel incredibly lucky to have had such a brilliant group ofresearchers to engage with throughout my doctoral studies. I would also like to thank Dr. EmilyTreleaven for her mentorship and for sharing her content area and methodological expertise withme and Dr. Josh Errickson for his statistical and methodological assistance. I am grateful for thementorship and guidance provided by Dr. William Lopez, who introduced me to the field ofPublic Health and the ways in which migration and health are intertwined. I am also verygrateful to the APHRC researchers and staff, including Dr. Martin Mutua, who welcomed meduring my brief stay in Nairobi and who were thoughtful and engaged resources as I developedmy dissertation and executed my research aims.I would also like to thank the Department of Epidemiology, including the faculty whohave instructed me over the past six years, Elvira Rivera and other department staff who havesupported me throughout my time with the department, my cohort, and my fellow BoultonResearch Group students.iii

I am forever grateful to my family and friends for supporting me throughout my studies.To my parents and my brother for your unwavering support, encouragement, and understandingas I navigated this program. To Brian, your support, advice, perspective, and positivity wereinstrumental to my progress through this program. And to Grace, Chelsea, Sarah, Freida, Chris,and many others for providing support and much-needed breaks from analysis and writing.Thank you, everyone.iv

Table of ContentsDedicationiiAcknowledgementsiiiList of TablesviList of FiguresixAbstractxiChapter 1 Introduction and Research Aims1Chapter 2 The Influence of Maternal Migration in Kenya: An Inverse Probability of TreatmentWeighted Analysis131Chapter 3 Childhood Vaccination Timeliness Following Maternal Migration to an InformalUrban Settlement in Kenya70Chapter 4 Migration to an Informal Urban Settlement and its Impact on Receipt of MaternalChildbirth Care in Nairobi, Kenya: An Exploration of Migrant Adaptation111Chapter 5 Discussion150v

List of TablesTable 1.1 Vaccination schedule under the Kenya Expanded Program on Immunization (2018) . 21Table 2.1 Kenya 2014 childhood vaccination schedule1,2 . 50Table 2.2 Descriptive statistics of children and their mothers included in the 2014 Kenya DHS,separately weighted by survey weights and IPTW-S weights . 51Table 2.3 Descriptive statistics of mothers’ migration status among women, separately weightedby survey weights and IPTS-S weights . 54Table 2.4 Balance diagnostics for IPTW weights . 55Table 2.5 Results of logistic regression examining the relationship between maternal migrationand FIC, among entire sample and stratified by migration stream . 56Table 2.6 Distributions of observed, imputed, and combined (observed imputed) up-to-datevaccination variable in 15 randomly selected imputations . 57Table 2.7 Characteristics of individuals with and without vaccination dates 1 . 58Table 2.8 Results of logistic regressions examining the relationships between maternal migrationand UTD, among entire sample and stratified by migration stream . 60Table 2.9 Sensitivity analyses comparing results of imputed analysis with complete case analysisand setting all children with missing vaccination cards to be UTD or not UTD 1 . 61Table 2.10 Balance diagnostics for IPTW weights for migration stream sensitivity analyses inwhich non-migrants in destination are set as referent group . 62vi

Table 2.11 Results of sensitivity analysis logistic regressions examining relationship betweenmaternal migration and FIC stratified by migration stream, with non-migrants in destination setas reference group . 63Table 2.12 E-values to assess the potential impact of unmeasured confounding . 64Table 3.1 Kenya childhood vaccination schedule (2018) . 91Table 3.2 Descriptive statistics of children and mothers included in the NUHDSS by maternalmigration status, unweighted and weighted by migrant status IPTW weight. 92Table 3.3 Descriptive statistics of mothers’ migration status (unweighted) . 94Table 3.4 Descriptive statistics of childhood vaccine receipt, separately among migrants and nonmigrants (unweighted) . 95Table 3.5 Balance diagnostics for IPTW weights . 96Table 3.6 Results of accelerated failure time models, comparing migrants vs. non-migrants,urban-origin vs. rural origin migrants, and first-time vs. circular migrants . 97Table 3.7 Sensitivity analysis of accelerated failure time models, comparing women whomigrated to join family vs. for 'other' reason, and comparing migrants vs. non-migrants whenchanging definition of migrant to moved within 5 years before childbirth . 99Table 3.8 Sensitivity analysis results of accelerated failure time models, comparing migrants vs.non-migrants, urban-origin vs. rural origin migrants, and first-time vs. circular migrants, amongfirstborn children . 100Table 4.1 Descriptive statistics of women's sociodemographic characteristics, NUHDSS 20042018. 132Table 4.2 Descriptive statistics of women’s migration characteristics, NUHDSS 2004-2018 . 133vii

Table 4.3 Results of logistic regressions examining the relationship between time spent in theinformal urban settlements and use of recommended childbirth services, in the entire sample andstratified by select variables . 134Table 4.4 Average marginal effect of adaptation, in the entire sample and stratified by selectvariables . 136Table 4.5 Results of examination of functional form of the number of years between migrationand childbirth . 137Table 4.6 Results of logistic regressions examining the relationship between time spent in theinformal urban settlements and use of recommended childbirth services, in the entire sample andstratified by select variables, among women's first childbirth . 138Table 4.7 Average marginal effect of adaptation, in the entire sample and stratified by selectvariables, among women's first childbirth . 140Table 4.8 Results of logistic regressions examining the relationship between time spent in theinformal urban settlements and use of recommended childbirth services stratified by maternitycare policy environment when childbirth occurred . 141viii

List of FiguresFigure 1.1 Map of 47 Kenyan counties color-coded by previous regions . 22Figure 1.2 Population distribution map of Kenya . 23Figure 1.3 Administrative maps of Kenya and Nairobi . 24Figure 1.4 Conceptual model of dissertation . 25Figure 2.1 Directed Acyclic Graph demonstrating theorized relationships between migration andvaccination . 65Figure 2.2 Figures of e-values for migrant status and migration stream analyses . 66Figure 3.1 Inverse survival curves comparing proportion of children who received each vaccinedose by child's age. Vertical lines indicate the age window during which vaccine receipt isrecommended. . 102Figure 3.2 Inverse survival curves comparing proportion of children who received each vaccinedose by child's age from sensitivity analysis in which only firstborn children were included.Vertical lines indicate the age window during which vaccine receipt is recommended. . 105Figure 4.1 Proportion of women receiving recommended childbirth care by number of yearsbetween migration and childbirth . 143Figure 4.2 Adjusted predicted probabilities of receiving recommended childbirth care by numberof years between in-migration and childbirth . 144Figure 4.3 Adjusted predicted probabilities of receiving recommended childbirth care by numberof years between in-migration and childbirth, among women's first childbirth . 145ix

Figure 4.4 Adjusted predicted probability of receiving recommended childbirth care by thenumber of years between in-migration and childbirth stratified by maternity care policyenvironment when the childbirth occurred, (a) among full sample and (b) among first birth . 146x

AbstractKenya has achieved impressive declines in maternal and under-five mortality over thelast few decades; maternal mortality has dropped from 687 to 342 deaths per 100,000 live birthsand under-five mortality has decreased from 104 to 43 deaths per 1,000 live births. However,accelerated progress will be necessary if Kenya is to reach the Sustainable Development Goaltargets of fewer than 70 maternal deaths per 100,000 live births and 25 under-five deaths per1,000 live births by 2030. As many of these deaths could be prevented with access to healthcareduring childbirth and early childhood, identifying factors leading to underutilization of care is akey strategy to reducing mortality. Research in low- and middle-income countries suggestsinternal migrants may be a particularly vulnerable group as the process of migration is disruptiveand typically requires a period of adaptation before women can effectively engage with thehealthcare system. This dissertation investigated the influence of maternal migration on the useof maternal and child healthcare services in Kenya.The first aim analyzed the relationship between maternal migration and receipt ofrecommended childhood vaccinations using nationally representative data from the 2014 KenyaDemographic and Health Survey. Migration status and migration stream (e.g., rural-urban) wereused as exposures and two measures of vaccination status, full and up-to-date vaccination, wereexplored as outcomes. After accounting for selection and confounding biases using multipleimputation and inverse probability of treatment weighting (IPTW), relationships betweenmigration and vaccination were statistically insignificant. These findings are an importantxi

deviation from previous literature that did not rigorously address important biases common tothis area of research.The second aim examined how maternal migration into informal urban settlements(IUSs) in Nairobi, Kenya influenced childhood vaccination timeliness. This aim leveraged 20022018 data from the Nairobi Urban Health and Demographic Surveillance System (NUHDSS).The primary analysis explored the impact of migration status and secondary analyses of migrantsexamined whether migrant origin or previous experience living in the IUS differentiallyinfluenced timely vaccine receipt. There was no evidence that migration status or characteristicsinfluenced vaccination timeliness in IPTW-weighted models. However, a considerable portion ofboth migrant and non-migrant children in the IUSs received their vaccinations late or not at all,indicating vaccination programs in the settlements should shift focus from simply increasingcoverage to improving timeliness.Using 2004-2018 NUHDSS data, the third aim analyzed the relationship between migrantwomen’s adaptation to living in an IUS and use of recommended childbirth services.Heterogeneity in the relationship between adaptation and childbirth care was explored bycharacteristics of the migration experience. Use of recommended childbirth services waswidespread in the IUSs but not associated with migrant adaptation. The relationship betweenadaptation and childbirth care did not differ significantly by a migration type, migration stream,migration companions, or reason for migrating.Collectively, these dissertation aims provide an in-depth analysis of the relationshipbetween migration and utilization of maternal and child healthcare services in Kenya. Findingssuggest that, in Kenya, characteristics enabling migration such as wealth and education, ratherthan the process of migration itself, drive differential healthcare use between migrants and non-xii

migrants. As the public health community works towards further global reductions in maternaland under-five mortality the populations of women and children who don’t receive adequatehealthcare must be clearly defined and targeted by outreach efforts.xiii

Chapter 1 Introduction and Research AimsAlthough the United Nations and the global public health community have identifiedmaternal and child health as key priorities, substantial international health disparities in maternaland child mortality persist.1–3 Nearly 94% of maternal deaths globally are among women livingin low- and middle-income countries with approximately two-thirds of those deaths in women insub-Saharan Africa.2 Similar global disparities exist in child mortality; children living in subSaharan Africa are 15 times more likely to die before their fifth birthday than children living inhigh-income countries.4 Many of these deaths could be prevented with access to proper medicalcare during childbirth and early childhood. In 2015, in an effort to improve the lives ofindividuals across the globe, the United Nations drafted the Sustainable Development Goals(SDGs).5 The SDGs were developed as a follow-up to the Millennium Development Goals(MDGs) to build upon the MDGs’ successes in improving global health.5 Importantly, the SDGswere developed through a consensus process that included more involvement from low- andmiddle-income countries than was present in the drafting of the MDGs.5 The SDGs provide aframework for global targets aimed at achieving important development goals ranging fromalleviating (or eliminating) poverty to increases in clean energy to health and wellbeing.5 SDGtargets 3.1 and 3.2 are particularly relevant to maternal and child health, aiming to decrease theglobal maternal mortality ratio to 70 deaths per 100,000 live births (and no country above 140deaths per 100,000 live births) and under-five mortality to 25 deaths per 1,000 live births by1

2030.1 Receipt of high quality health services throughout childbirth and childhood is critical toimproving maternal and child health and eliminating preventable mortality.2,4For women, many complications of childbirth, such as infection and severe bleeding, canbe deadly without the assistance of skilled healthcare providers in a clinical setting withappropriate supplies. Yet, childbirth without a skilled attendant and outside a health facility isstill common in many low- and middle-income countries (LMICs).2 Among children, vaccinepreventable diseases, such as pneumonia and diarrhea, are leading causes of death despite theprovision of recommended vaccines for free in most LMICs under the Expanded Program onImmunization.4 In an effort to decrease disparities in maternal and child mortality it is imperativeto understand the factors contributing to the underutilization of these lifesaving maternal andchild healthcare services. Historically the literature has explored the influence of family features,like wealth,6–10 and maternal characteristics, like education and empowerment,9–12 as drivers ofmaternal and child healthcare use. Another potential factor may be maternal migration, asresearch suggests migration influences a variety of indicators of health and healthcare utilization,including under-five mortality,13 incomplete childhood immunization,14 and antenatal care andhealth facility delivery.15This dissertation explored the influence of maternal migration on the use of maternal andchild healthcare services in Kenya. The first aim investigated the relationship between maternalmigration and the receipt of recommended childhood vaccinations using nationallyrepresentative data from the 2014 Kenya Demographic and Health Survey. The second and thirdaims examined how maternal migration into two informal urban settlements in Nairobi, Kenyaimpacted vaccination timeliness (Aim 2) and the use of childbirth services (Aim 3) using 20022018 data from the Nairobi Urban Health and Demographic Surveillance System. This2

introductory chapter discusses the country of Kenya and outlines key concepts relevant tochildhood vaccination, childbirth services, migration and health, and the informal urbansettlements of Nairobi.KenyaWith a population of approximately 54.6 million, people Kenya, located in EasternAfrica, is among the top ten most populated countries on the African continent.16 It hasdemonstrated consistently strong economic growth (5%-6% real gross domestic product growthrate since 2010) and transitioned from a low- to a middle-income economy in 2014.16,17 ThoughKenya’s economy is still primarily agrarian, its tourism and technology sectors are booming,making Kenya an important economic, technology, and travel hotspot in Africa.16,18 Thisgrowing economy has led to substantial increases in the quality of life of Kenyan residents.Kenya’s human development index (HDI, a measure that combines national-level lifeexpectancy, education, and income levels into a number ranging from 0 to 1) increased from 0.46to 0.60 between 2000 and 2019.19 However, when the level of inequality is accounted for theIHDI (inequality-adjusted HDI) shrinks to 0.44, representing a loss in development of 26% dueto intranational inequalities in life expectancy, education, and income.19 Kenya’s Gini index (ameasure of income inequality in which higher values indicate more inequality) similarlydemonstrates sizable income inequality in Kenya. At 40.8, Kenya’s 2015 Gini index was not ashigh as countries like South Africa (63.0), but was higher than other countries in eastern Africaincluding Tanzania (37.8) and Ethiopia (35.0).20 These inequalities are particularly stark whencomparing Kenya’s rural and urban populations. Urban dwellers enjoy substantially highereducation (22% of men and 19% of women have received higher than a secondary education)3

and more wealth (49% of urban households fall into the richest wealth quintile) than ruralpopulations, of whom only 7% and 6% of men and women, respectively have higher than asecondary education and a mere 5% of households belong to the richest wealth quintile. 21Kenya’s government is a presidential republic.16 Prior to 2010, political power in Kenyawas highly centralized with an unicameral National Assembly and a president who wieldedimmense authority including the ability to, at will, dismiss judges and detain citizens without atrial.16,22 In a 2010 national referendum Kenyans voted to approve a new Constitution, whichmoved power and resources away from the federal government, reinstated a bicameralparliament, created a Supreme Court, and dissolved the eight regions (Central, Coast, Eastern,Nairobi, North Eastern, Nyanza, Rift Valley, and Western) in favor of 47 counties (Figure1.1).16,22 This new governmental system was incrementally phased-in beginning with 2013county Governor elections and aiming for the new decentralized government to obtain fullauthority by 2016.22Following this decentralization of political power, in 2013 Kenya’s health system wasalso restructured and much of the authority and responsibilities for health services moved fromthe central government to the local county governments.21,23 The goal of this devolution of thehealth sector was to promote efficiency and equality in health services and reduce regionaldisparities in health outcomes.23 As an example of the stark disparities in healthcare accessbefore devolution, in 2008-2009 96.4% of women living in the Nairobi region received antenatalcare from a skilled provider (doctor, nurse, midwife, etc.) and 89.4% gave birth in a healthfacility while in the North Eastern region only 69.5% of women received skilled antenatal careand only 17.3% of women delivered their child in a health facility.24 Childhood vaccinationcoverage demonstrated similar regional inequities, with nearly 86% of children in the Central4

region receiving all basic immunizations compared to less than 50% of children in the NorthEastern region.24 After decentralization of the health sector, the central government now createsnational health policy and provides technical and capacity building support to counties, whilecounty governments are responsible for the provision of public health services and the promotionof health services utilization, including the organization of community health volunteers andcommunity health extension workers.21,23 The central government of Kenya has developed avariety of maternal and child health programs and policies designed to improve maternal andchild health and increase health service use such as the Kenya Expanded Program onImmunization, free delivery services at public health facilities, the Integrated Management ofChildhood Illness Initiative, the Community-Based Newborn Care Program, the Infant andYoung Child Feeding Program, Malezi Bora (maternal and child health and nutrition weeks heldtwice a year), and campaigns to increase the receipt of vitamin A supplementation anddeworming medications.21,25Child vaccination in KenyaChildhood mortality in Kenya has seen a striking decrease since the early 2000s, withunder-five mortality dropping from 100 to 43 deaths per 1,000 live births between 1990 and2019, though reaching the 2030 SDG target of 25 under-five deaths per 1,000 live births hasremained elusive.3,26 This decline in under-five mortality is largely the result of Kenya’simpressive improvements in vaccination coverage since the introduction of the Kenya ExpandedProgram on Immunization (KEPI), as 35% of under-five deaths in Kenya are due to diarrhealdiseases, pneumonia, and measles, all of which are vaccine preventable. 27,28 The KEPI wasinstituted in 1980 with the goal of providing vaccines that protect against tuberculosis, polio,5

diphtheria, pertussis, tetanus, and measles (the “basic” vaccination series) to the children ofKenya for free.27 Since its initiation, new vaccines have been added to the KEPI schedule whichnow includes 15 doses: one dose of Bacille Calmette- Guérin vaccine given at birth (BCG,protects against tuberculosis); three doses each of the oral poliovirus vaccine (OPV), thepentavalent vaccine (penta, includes diphtheria, tetanus, pertussis [DTP], hepatitis B [HBV], andHaemophilus influenza type b [Hib] antigens), and the pneumococcal conjugate vaccine (PCV)given at 6, 10, and 14 weeks; one dose of the inactivated poliovirus vaccine (IPV) given at 10weeks; two doses of the rotavirus vaccine (rota) given at 6 and 10 weeks, and two doses of themeasles vaccine given at 9 and 18 months (Table 1). 21,29Implementation of the KEPI has resulted in significant increases in vaccination coveragein Kenya. Coverage of the third dose of the diphtheria-tetanus-pertussis-containing vaccine(DPT3, a precursor to the pentavalent vaccine and an indicator often used to estimate vaccinationprogram performance) among children aged 12-23 months increased from 58% in 1984 to 92%in 2019.27,30 Other doses followed a similar trend with coverage of BCG increasing from 76% to95% and the 9-month dose of the measles vaccine increasing from 55% to 89% in the sameperiod.30 The proportion of children who are fully vaccinated with all recommended doses hasfollowed a similar trend. Further, receipt of all vaccines (except PCV, rota, and IPV, whichweren’t introduced until later) has continued to increase consistently over the past two decadesfrom 57% in 2003 to 77% in 2008-2009 and 79% in 2014.21 These advances are considerableand receipt of many individual doses reach the World Health Organization Africa RegionalStrategic Plan for Immunization’s goal of 90% national coverage in by 2020.31 However, DPT3coverage at the county level demonstrates less success as almost half of Kenya’s 47 countieshave DPT3 coverage below 80%, falling below the Plan’s goal of 80% coverage in all districts6

and representing important geographic disparities in vaccination coverage.31,32 Clearly, muchprogress has been made in providing immunization services to children in Kenya, but much workremains to ensure this life-saving service is available to all Kenyan children.Estimating cumulative vaccination coverage at certain age intervals (e.g., proportion ofchildren aged 12-23 months who have received DPT3) is the most straightforward and commonmanner in which researchers and public health officials estimate community protection againstvaccine preventable diseases (VPDs). However, simple coverage estimates can mask importantdelays in vaccine receipt that may result in substantial disease transmission among children whoare younger than the examined age interval and at ages when they are particularly prone to moreserious disease, hospitalization, and death. These delays leave children susceptible to VPDslonger than necessary and place communities at risk of VPD outbreaks.33,34 In an effort to betterunderstand child and community protection against VPDs researchers have begun examiningvaccination timeliness in addition to vaccination coverage. Vaccination timeliness, calculated asthe continuous measure of a child’s age at vaccine receipt, can provide a better characterizationof whether a child is delayed in receiving individual vaccines and whether a community hassufficient vaccination levels to protect against outbreaks. Using this continuous measure of age atvaccination, a particular vaccine dose is considered timely if it is received within one month ofthe date recommended by a country’s national vaccination schedule.35 Interest in estimatingvaccination timeliness has increased dramatically in the last ten years as vaccination programstransition from simply optimizing vaccination coverage to ensuring timely vaccine receipt. Theneed for an increased focus on vaccine timeliness is exemplified by Kenya, which has high levelsof vaccine coverage but only 37% of children currently aged 12-36 months received all eightbasic KEPI vaccines on time.8 This proportion masks a significant inequity in timeliness by7

urbanicity, with 48% of urban children compared to only 31% of rural children receiving allbasic vaccines on time.8 Similar to vaccination coverage, there is much work to be done todecrease disparities in vaccination timeliness in Kenya and ensure all children not only receiveeach recommended dose but receive each dose on time.Maternal childbirth services in KenyaKenya has also achieved impressive improvements in maternal mortality in the 2000swith the maternal mortality ratio declining from 678 to 342 deaths per 100,000 live birthsbetween 2003 and 2017.36 Though still well above the 2030 SDG target of 70 deaths per 100,000live births, it is clear that decreasing maternal mortality has been

indicating vaccination programs in the settlements should shift focus from simply increasing coverage to improving timeliness. Using 2004-2018 NUHDSS data, the third aim analyzed the relationship between migrant women's adaptation to living in an IUS and use of recommended childbirth services.